issues
search
bojone
/
rerope
Rectified Rotary Position Embeddings
330
stars
27
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Generating same token
#21
Madhu000
opened
3 months ago
0
"position_ids_q - position_ids_k < window" without abs()
#20
wzhcz8902
opened
4 months ago
0
position_ids_q - position_ids_k < window
#19
wzhcz8902
closed
4 months ago
0
[BUG] rope implementation question
#18
wzhcz8902
closed
4 months ago
1
[BUG] relative position ids seems wrong
#17
wzhcz8902
closed
4 months ago
2
cos and sin dimension
#16
wzhcz8902
closed
4 months ago
1
<q_len=1> question
#15
wzhcz8902
closed
4 months ago
1
LlamaLinearScalingRotaryEmbedding接口参数问题
#14
wzhcz8902
closed
4 months ago
1
Dataset of ReROPE eval
#13
Madhu000
closed
3 months ago
1
数据集samples_15k
#12
zkcpku
closed
4 months ago
2
请问在推理阶段,为什么需要对input部份计算注意力呢?
#11
Anditty
opened
10 months ago
1
为什么ntk rope那里要乘以-a
#10
linyubupa
closed
10 months ago
1
typo(project): add requirements.txt
#9
tpoisonooo
closed
11 months ago
0
problem with rerope_patch
#8
Arist12
opened
1 year ago
1
Blogs in English
#7
NormXU
closed
1 year ago
3
运行 test.py 显存爆了
#6
liyi-ff
opened
1 year ago
3
ntk_rope_mixed_init 中old_init是否可以简化,省略inv_freq、_set_cos_sin_cache()步骤
#5
samantha0-ops
opened
1 year ago
3
Monkey patch for original LLaMA2 code.
#4
WeixuanXiong
opened
1 year ago
1
rerope:为什么对大于train_seqlength的query token要稍微放大一点
#3
Liu20210916
opened
1 year ago
2
测了一下千问
#2
af-74413592
opened
1 year ago
5
Excellent Idea!!!!
#1
0three
opened
1 year ago
3